agentv 4.22.0 → 4.23.0-next.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/{artifact-writer-S7S6K3XA.js → artifact-writer-KBYDJNKS.js} +4 -4
- package/dist/{chunk-5QF7PKZ4.js → chunk-32VOBCSX.js} +21 -25
- package/dist/chunk-32VOBCSX.js.map +1 -0
- package/dist/{chunk-GMMIWIBF.js → chunk-4ISLX6RP.js} +12 -11
- package/dist/chunk-4ISLX6RP.js.map +1 -0
- package/dist/{chunk-4YZDPAGZ.js → chunk-KX2MJDMY.js} +73 -29
- package/dist/chunk-KX2MJDMY.js.map +1 -0
- package/dist/{chunk-762UZSGR.js → chunk-SBI4Z7YY.js} +8 -9
- package/dist/chunk-SBI4Z7YY.js.map +1 -0
- package/dist/{chunk-CM2AVQAM.js → chunk-ZVEESBXF.js} +7 -5
- package/dist/chunk-ZVEESBXF.js.map +1 -0
- package/dist/cli.js +5 -5
- package/dist/{dist-6LNJDZAI.js → dist-A7D52JMJ.js} +5 -3
- package/dist/index.js +5 -5
- package/dist/{interactive-7LC5WFP3.js → interactive-66Z7DNMJ.js} +5 -5
- package/dist/{ts-eval-loader-PA4YFM5D-2DKQQQOZ.js → ts-eval-loader-HT7NCO4M-V63KENQZ.js} +2 -2
- package/package.json +1 -1
- package/dist/chunk-4YZDPAGZ.js.map +0 -1
- package/dist/chunk-5QF7PKZ4.js.map +0 -1
- package/dist/chunk-762UZSGR.js.map +0 -1
- package/dist/chunk-CM2AVQAM.js.map +0 -1
- package/dist/chunk-GMMIWIBF.js.map +0 -1
- /package/dist/{artifact-writer-S7S6K3XA.js.map → artifact-writer-KBYDJNKS.js.map} +0 -0
- /package/dist/{dist-6LNJDZAI.js.map → dist-A7D52JMJ.js.map} +0 -0
- /package/dist/{interactive-7LC5WFP3.js.map → interactive-66Z7DNMJ.js.map} +0 -0
- /package/dist/{ts-eval-loader-PA4YFM5D-2DKQQQOZ.js.map → ts-eval-loader-HT7NCO4M-V63KENQZ.js.map} +0 -0
|
@@ -13,11 +13,11 @@ import {
|
|
|
13
13
|
writeArtifacts,
|
|
14
14
|
writeArtifactsFromResults,
|
|
15
15
|
writePerTestArtifacts
|
|
16
|
-
} from "./chunk-
|
|
17
|
-
import "./chunk-
|
|
16
|
+
} from "./chunk-ZVEESBXF.js";
|
|
17
|
+
import "./chunk-SBI4Z7YY.js";
|
|
18
18
|
import "./chunk-QOBQ5XYF.js";
|
|
19
19
|
import "./chunk-BPGJ4HBU.js";
|
|
20
|
-
import "./chunk-
|
|
20
|
+
import "./chunk-KX2MJDMY.js";
|
|
21
21
|
import "./chunk-ZKO2LGRR.js";
|
|
22
22
|
import "./chunk-NPVGBFF6.js";
|
|
23
23
|
import "./chunk-HQDCIXVH.js";
|
|
@@ -38,4 +38,4 @@ export {
|
|
|
38
38
|
writeArtifactsFromResults,
|
|
39
39
|
writePerTestArtifacts
|
|
40
40
|
};
|
|
41
|
-
//# sourceMappingURL=artifact-writer-
|
|
41
|
+
//# sourceMappingURL=artifact-writer-KBYDJNKS.js.map
|
|
@@ -15,7 +15,7 @@ import {
|
|
|
15
15
|
resolveWorkspaceOrFilePath,
|
|
16
16
|
toSnakeCaseDeep,
|
|
17
17
|
writeArtifactsFromResults
|
|
18
|
-
} from "./chunk-
|
|
18
|
+
} from "./chunk-ZVEESBXF.js";
|
|
19
19
|
import {
|
|
20
20
|
ResponseCache,
|
|
21
21
|
RunBudgetTracker,
|
|
@@ -31,7 +31,7 @@ import {
|
|
|
31
31
|
shouldSkipCacheForTemperature,
|
|
32
32
|
stageResultsArtifacts,
|
|
33
33
|
syncResultsRepo
|
|
34
|
-
} from "./chunk-
|
|
34
|
+
} from "./chunk-SBI4Z7YY.js";
|
|
35
35
|
import {
|
|
36
36
|
CLI_PLACEHOLDERS,
|
|
37
37
|
COMMON_TARGET_SETTINGS,
|
|
@@ -52,6 +52,7 @@ import {
|
|
|
52
52
|
loadConfig,
|
|
53
53
|
loadTestSuite,
|
|
54
54
|
normalizeLineEndings,
|
|
55
|
+
parseYamlValue,
|
|
55
56
|
readTargetDefinitions,
|
|
56
57
|
readTestSuiteMetadata,
|
|
57
58
|
resolveFileReference,
|
|
@@ -62,7 +63,7 @@ import {
|
|
|
62
63
|
subscribeToCopilotSdkLogEntries,
|
|
63
64
|
subscribeToPiLogEntries,
|
|
64
65
|
toCamelCaseDeep
|
|
65
|
-
} from "./chunk-
|
|
66
|
+
} from "./chunk-KX2MJDMY.js";
|
|
66
67
|
|
|
67
68
|
// src/commands/eval/shared.ts
|
|
68
69
|
import { constants } from "node:fs";
|
|
@@ -178,7 +179,7 @@ async function findRepoRoot(start) {
|
|
|
178
179
|
// package.json
|
|
179
180
|
var package_default = {
|
|
180
181
|
name: "agentv",
|
|
181
|
-
version: "4.
|
|
182
|
+
version: "4.23.0-next.1",
|
|
182
183
|
description: "CLI entry point for AgentV",
|
|
183
184
|
type: "module",
|
|
184
185
|
repository: {
|
|
@@ -2765,31 +2766,24 @@ function formatMatrixSummary(results) {
|
|
|
2765
2766
|
// ../../packages/core/dist/evaluation/validation/index.js
|
|
2766
2767
|
import { readFile as readFile2 } from "node:fs/promises";
|
|
2767
2768
|
import path14 from "node:path";
|
|
2768
|
-
import { parse } from "yaml";
|
|
2769
2769
|
import { readFile as readFile22, readdir, stat as stat2 } from "node:fs/promises";
|
|
2770
2770
|
import path22 from "node:path";
|
|
2771
|
-
import { parse as parse2 } from "yaml";
|
|
2772
2771
|
import { readFile as readFile3 } from "node:fs/promises";
|
|
2773
2772
|
import path32 from "node:path";
|
|
2774
|
-
import { parse as parse3 } from "yaml";
|
|
2775
2773
|
import { readFile as readFile4 } from "node:fs/promises";
|
|
2776
2774
|
import path42 from "node:path";
|
|
2777
|
-
import { parse as parse4 } from "yaml";
|
|
2778
2775
|
import { readFile as readFile5 } from "node:fs/promises";
|
|
2779
|
-
import { parse as parse5 } from "yaml";
|
|
2780
2776
|
import { readFile as readFile6 } from "node:fs/promises";
|
|
2781
2777
|
import path52 from "node:path";
|
|
2782
|
-
import { parse as parse6 } from "yaml";
|
|
2783
2778
|
import { access as access4, readFile as readFile7 } from "node:fs/promises";
|
|
2784
2779
|
import path62 from "node:path";
|
|
2785
|
-
import { parse as parse7 } from "yaml";
|
|
2786
2780
|
var SCHEMA_EVAL_V2 = "agentv-eval-v2";
|
|
2787
2781
|
var SCHEMA_TARGETS_V2 = "agentv-targets-v2.2";
|
|
2788
2782
|
var SCHEMA_CONFIG_V2 = "agentv-config-v2";
|
|
2789
2783
|
async function detectFileType(filePath) {
|
|
2790
2784
|
try {
|
|
2791
2785
|
const content = await readFile2(filePath, "utf8");
|
|
2792
|
-
const parsed =
|
|
2786
|
+
const parsed = parseYamlValue(content);
|
|
2793
2787
|
if (Array.isArray(parsed)) {
|
|
2794
2788
|
return "cases";
|
|
2795
2789
|
}
|
|
@@ -2865,7 +2859,9 @@ var KNOWN_TOP_LEVEL_FIELDS = /* @__PURE__ */ new Set([
|
|
|
2865
2859
|
"assertions",
|
|
2866
2860
|
"evaluators",
|
|
2867
2861
|
"preprocessors",
|
|
2868
|
-
"workspace"
|
|
2862
|
+
"workspace",
|
|
2863
|
+
"metadata",
|
|
2864
|
+
"governance"
|
|
2869
2865
|
]);
|
|
2870
2866
|
var DEPRECATED_TOP_LEVEL_FIELDS = /* @__PURE__ */ new Map([
|
|
2871
2867
|
["eval_cases", "'eval_cases' is deprecated. Use 'tests' instead."],
|
|
@@ -2940,7 +2936,7 @@ async function validateEvalFile(filePath) {
|
|
|
2940
2936
|
let parsed;
|
|
2941
2937
|
try {
|
|
2942
2938
|
const content = await readFile22(absolutePath, "utf8");
|
|
2943
|
-
parsed = interpolateEnv(
|
|
2939
|
+
parsed = interpolateEnv(parseYamlValue(content), process.env);
|
|
2944
2940
|
} catch (error) {
|
|
2945
2941
|
errors.push({
|
|
2946
2942
|
severity: "error",
|
|
@@ -3216,7 +3212,7 @@ async function validateWorkspaceConfig(workspace, evalFilePath, errors, location
|
|
|
3216
3212
|
const workspacePath = path22.resolve(path22.dirname(evalFilePath), workspace);
|
|
3217
3213
|
try {
|
|
3218
3214
|
const workspaceContent = await readFile22(workspacePath, "utf8");
|
|
3219
|
-
const parsedWorkspace = interpolateEnv(
|
|
3215
|
+
const parsedWorkspace = interpolateEnv(parseYamlValue(workspaceContent), process.env);
|
|
3220
3216
|
if (!isObject(parsedWorkspace)) {
|
|
3221
3217
|
errors.push({
|
|
3222
3218
|
severity: "error",
|
|
@@ -3618,7 +3614,7 @@ async function validateCasesFile(filePath) {
|
|
|
3618
3614
|
let parsed;
|
|
3619
3615
|
try {
|
|
3620
3616
|
const content = await readFile3(absolutePath, "utf8");
|
|
3621
|
-
parsed =
|
|
3617
|
+
parsed = parseYamlValue(content);
|
|
3622
3618
|
} catch (error) {
|
|
3623
3619
|
errors.push({
|
|
3624
3620
|
severity: "error",
|
|
@@ -3907,7 +3903,7 @@ async function validateTargetsFile(filePath) {
|
|
|
3907
3903
|
let parsed;
|
|
3908
3904
|
try {
|
|
3909
3905
|
const content = await readFile4(absolutePath, "utf8");
|
|
3910
|
-
parsed =
|
|
3906
|
+
parsed = parseYamlValue(content);
|
|
3911
3907
|
} catch (error) {
|
|
3912
3908
|
errors.push({
|
|
3913
3909
|
severity: "error",
|
|
@@ -4122,7 +4118,7 @@ async function validateConfigFile(filePath) {
|
|
|
4122
4118
|
const errors = [];
|
|
4123
4119
|
try {
|
|
4124
4120
|
const content = await readFile5(filePath, "utf8");
|
|
4125
|
-
const parsed =
|
|
4121
|
+
const parsed = parseYamlValue(content);
|
|
4126
4122
|
if (typeof parsed !== "object" || parsed === null) {
|
|
4127
4123
|
errors.push({
|
|
4128
4124
|
severity: "error",
|
|
@@ -4275,7 +4271,7 @@ async function validateFileReferences(evalFilePath) {
|
|
|
4275
4271
|
let parsed;
|
|
4276
4272
|
try {
|
|
4277
4273
|
const content = await readFile6(absolutePath, "utf8");
|
|
4278
|
-
parsed =
|
|
4274
|
+
parsed = parseYamlValue(content);
|
|
4279
4275
|
} catch {
|
|
4280
4276
|
return errors;
|
|
4281
4277
|
}
|
|
@@ -4393,7 +4389,7 @@ async function validateWorkspacePaths(evalFilePath) {
|
|
|
4393
4389
|
let parsed;
|
|
4394
4390
|
try {
|
|
4395
4391
|
const content = await readFile7(absolutePath, "utf8");
|
|
4396
|
-
parsed =
|
|
4392
|
+
parsed = parseYamlValue(content);
|
|
4397
4393
|
} catch {
|
|
4398
4394
|
return errors;
|
|
4399
4395
|
}
|
|
@@ -4404,7 +4400,7 @@ async function validateWorkspacePaths(evalFilePath) {
|
|
|
4404
4400
|
const workspaceFilePath = path62.resolve(evalDir, workspaceRaw);
|
|
4405
4401
|
try {
|
|
4406
4402
|
const wsContent = await readFile7(workspaceFilePath, "utf8");
|
|
4407
|
-
const wsParsed =
|
|
4403
|
+
const wsParsed = parseYamlValue(wsContent);
|
|
4408
4404
|
if (isObject5(wsParsed)) {
|
|
4409
4405
|
const wsDir = path62.dirname(workspaceFilePath);
|
|
4410
4406
|
await validateWorkspaceObject(wsParsed, wsDir, absolutePath, "workspace", errors);
|
|
@@ -5435,7 +5431,7 @@ async function runEvalCommand(input) {
|
|
|
5435
5431
|
const useFileExport = !!options.otelFile;
|
|
5436
5432
|
if (options.exportOtel || useFileExport) {
|
|
5437
5433
|
try {
|
|
5438
|
-
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-
|
|
5434
|
+
const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-A7D52JMJ.js");
|
|
5439
5435
|
let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
|
|
5440
5436
|
let headers = {};
|
|
5441
5437
|
if (options.otelBackend) {
|
|
@@ -5634,7 +5630,7 @@ async function runEvalCommand(input) {
|
|
|
5634
5630
|
const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
|
|
5635
5631
|
let transcriptProviderFactory;
|
|
5636
5632
|
if (options.transcript) {
|
|
5637
|
-
const { TranscriptProvider } = await import("./dist-
|
|
5633
|
+
const { TranscriptProvider } = await import("./dist-A7D52JMJ.js");
|
|
5638
5634
|
const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
|
|
5639
5635
|
const totalTests = [...fileMetadata.values()].reduce(
|
|
5640
5636
|
(sum, meta) => sum + meta.testCases.length,
|
|
@@ -5803,7 +5799,7 @@ async function runEvalCommand(input) {
|
|
|
5803
5799
|
if (usesDefaultArtifactWorkspace && allResults.length > 0) {
|
|
5804
5800
|
const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
|
|
5805
5801
|
if (isResumeAppend) {
|
|
5806
|
-
const { writePerTestArtifacts } = await import("./artifact-writer-
|
|
5802
|
+
const { writePerTestArtifacts } = await import("./artifact-writer-KBYDJNKS.js");
|
|
5807
5803
|
await writePerTestArtifacts(allResults, runDir, {
|
|
5808
5804
|
experiment: normalizeExperimentName(options.experiment)
|
|
5809
5805
|
});
|
|
@@ -6026,4 +6022,4 @@ export {
|
|
|
6026
6022
|
getCategories,
|
|
6027
6023
|
filterByCategory
|
|
6028
6024
|
};
|
|
6029
|
-
//# sourceMappingURL=chunk-
|
|
6025
|
+
//# sourceMappingURL=chunk-32VOBCSX.js.map
|